Convincing Human Thread Title Not Generated by AI

[Brian C]

I spent some time yesterday playing around with the AI tool we’ve been given at work, and it seemed pretty useless even though we’re encouraged to use it. I work in a pretty niche field, which makes it all the more odd that we’re being told to incorporate it into our workflow.

Basically, it would just talk itself in circles. It did give superficially reasonable answers to very basic questions, but I’d lose it whenever I tried to get even a little into the weeds.

ME: Does this document meet requirements?

AI: Yes, this document meets requirements because x, y, and z. Would you like me to draft an email approving this document?

ME: But the document is missing x and z, isn’t it? [even though doc actually had x!]

AI: That’s true. You are right to have concerns about this document. Would you like me to draft an email requesting x and z?

So my takeaway is that it’s basically an email drafting tool. It is definitely not remotely close to being able to do my job, but whether it’s close enough to make the execs think it could maybe do my job is of course an open question.

[knives]

As time goes by I see ‘AI’ as being just another black box like radiation and even CPUs used to be. People think it can do anything and trust it because they don’t understand how it works. I was just reading an article about a company that went bankrupt because it was having AI monitor efficiency and they just trusted it. Turns out it spent months just making up data because analyzing the data accurately wasn’t necessary for its task. Same thing with how that internet outage some months ago was caused by Amazon tasking AI with identifying and getting rid of bugs so the program just deleted code where the bugs were. It should have instead been used to send reports of bugs to people to fix.

AI is incredibly useful for what it is intended for, simplifying certain tasks and being a resource to hone your own ideas with, but otherwise useless.

[Brian C]

I think it’s begging the question to an extreme degree to say that it’s extremely useful “for what it’s intended for,” in all sorts of ways.

[The Curious Sofa]

I’m a translator specialising in museum texts, art books and exhibition catalogues. Recently, I was commissioned to translate wall texts and a catalogue for a science-focused exhibition on a subject well outside my usual remit. AI proved to be a valuable research tool, helping me identify the correct technical terminology, though its effectiveness depended heavily on how I framed the prompt. In the end pretty much all of the terminology and explanations were accurate (I double checked everything), and the final translation passed scientific review by the editor with flying colours.

The idea that AI can only draft emails may have been true a few years ago. The technology is clearly capable of far more and I’ve seen how much it has advanced in just the past two years and it will keep evolving. But like with any tool, it depends on how you use it. While it cannot yet fully replace my work, I am convinced it will not be long before it can. Fortunately, I am not too far from retirement.

[HJackson]

You're saying the technology is "clearly capable of far more" and then the application you described (and correct me if I'm underselling it in a really unfair way) sounds like a highly efficient Google search? Meanwhile we're being bombarded with propaganda constantly that it's capable of replacing large swathes of white collar work and will transform the world economy if we just build another ten data centres the size of Manhattan. The gulf between these positions is enormous.

[The Curious Sofa]

Isn't translation a white-collar job? My colleagues who translate manuals for washing machines and the likes are now reaching the end of their careers and will need to retrain in a different profession. The fact that I am more specialised still saves me for now, but I have even seen AI improving on a monthly basis. I used to work in animation, and many simpler tasks, such as in-betweening, can now be done by AI. Animation is also a white-collar job. Here in Germany there is currently a lot of commotion because it looks like the entire movie dubbing industry is going to be replaced by AI. These are white-collar jobs too. Just because your job isn't threatened, doesn't mean many others aren't.

[Drucker]

I think the idea that some (mostly entry-level, data-entry type) white collar tasks will be taken over by AI is fairly likely. I work in advertising sales (doing back-end work such as Business Intelligence and management reporting) and a decade ago I left a position in part because I was sure it could be automated.

But the idea that the general population will be engaging with these little agents to a degree to which they would disrupt the entire economy in the manner that has been described in the blog post I linked to is preposterous. People complain when their grocery bill goes up $10, and this article asks "what if people click a button that allows shopping to occur 24/7 for individual consumers, even when their phone is down." Preposterous.

[Peacock]

I've experimented with AI editing tools for cutting down interviews many times recently, but I find they basically function like an extremely stupid beginner Editing Assistant. So you end up having to micromanage every step they take and argue with them to stick to whatever framework you've both settled on etc, sometimes it ends up taking as much time as editing it all yourself... but usually it ends up taking a lot more time.

I don't consider this to be a flaw of these AI editing tools themselves but the OpenAI brain behind them. I remember Noam Chomsky being asked a few years ago about the great replacement theory and AI becoming a thread if it becomes too smart.. and him dismissing it as being so far from that that it's almost science fiction at this stage. And I agree... AI tools are sometimes great for very basic but laborious tasks but it isn't smart. If I put in an hour long interview transcript into AI and ask it to cut it into down into a certain template it cannot come close to competing with the human mind. It lacks any nuance or creative skill with tasks like this.

Jackson talks about using AI instead of Google, and yes that's something I find it useful for - for example if I need to identify a piece of music in an obscure Japanese film which isn't mentioned in the credits and no one on any English language forums discusses it, I can get AI to deep research Japanese language websites and find the answer. It's really useful when working with other languages you have no knowledge of.

Jumping back to video editing; a couple of colleagues I've met have also played with it for things like social media content work and agree that it's a long way off replacing an actual editor. But it may help speed up editing workflows. Just as AI masking tools now massively speed up complex keying work when grading.

I'm not suggesting that because of these examples that no one in the world is at risk of losing work due to AI, many people as mentioned by others in this thread, have lost their jobs or are at risk. And no one buys the idea that those lost jobs will be 1-1 replaced with equivalent AI related jobs.

But the genie can't be put back in the bottle... AI isn't going anywhere, and for many people I know it's been an adapt or die situation.

[MichaelB]

Google's AI summaries occasionally throw up something I might not have come across (or considered) before, but I always, always then research this the old-fashioned way. Partly because AI sometimes blatantly makes stuff up, but also because I want an information paper trail—and indeed on a recent film dictionary job I was required to provide three legitimate references (preferably books with ISBN numbers) for every entry that I wrote, regardless of how I stumbled upon the info in the first place. Similarly, foreign editions of Wikipedia can be a superb source of links and references—when working on commentaries for The Devil's Bride (Lithuania) and Dead Mountaineer's Hotel (Estonia), the local Wikipedias had tons of info—and links—that the English version unsurprisingly didn't come anywhere near.

But the fact that I take a similar approach to AI and non-AI sources should be revealing enough in itself.

Same with AI transcription—I use it primarily for two things: transcribing long interviews that I've been commissioned to cut down, and transcribing commentaries for rightsholders who insist on such things being sent to their legal departments. In the case of the former, they're purely an internal reference so mistakes don't matter. With the latter, I tweak them manually as I edit and mix the commentaries in question, but I don't need 100% accuracy as they're not intended for publication.

But if they were intended for publication, I might well still do it the old-fashioned way, as I find that cleaning them up is often fiddlier than transcribing them from scratch. Similarly, I don't use AI for subtitling at all; even aside from transcription niggles, it invariably places line/breaks where I'd/never consider placing/them, and this takes ages to rectify—typically much longer than simply creating the subtitles from scratch.

As with other revolutions—desktop publishing, the internet, etc.—AI is basically a tool, and one that can be a huge timesaver if you know what you're doing and you have a good idea of roughly how much time it will save.

[spectre]

One of my chief bugbears is the term itself, "AI". It's now popularly used to describe a wide range of tools – from LLMs to generated images to older-generation stuff like chatbots and NPCs in videogames – that aren't necessarily doing the same thing, and in many cases are just standard computing dealing with bigger loads of data. As "AI", all of this gets placed under an umbrella term that evokes science-fiction concepts about sentient robots, none of which are remotely anything like what's being achieved with these technologies.

In the recent past, we all had at least a vague concept of what "AI" was as a futuristic or speculative technology; now, a massive industry based on hype is using the same term to describe processes that look like they're intelligent but aren't in any meaningful way, which has in turn conned employers – who, as knives suggests, have no idea what this thing is or how it actually works – into thinking we're on the threshold of a brave new world where robots are able to take over (or even improve on) tasks that previously required human intelligence. The branding is a big part of that.

I don't have much else to add that hasn't already been articulated above, but I do feel that even calling this "AI" when criticising the industry is conceding part of the argument. So when, for instance, Paul Schrader fawningly posts a clip of the latest slop to emerge from Darren Aronofsky's studio and suggests it heralds the end of human-made cinema, it'd be nice for people to stop for a second and acknowledge what this actually is: a CGI animation made with automated image-generation technology and human voice actors, not an early work from the Skynet Cinematic Universe. The development of photorealistic animation (which this isn’t, but is arguably approaching) is interesting in its own right, but it seems it’s yet another thing that’s been squeezed under the AI hype umbrella. At least calling all this what it is might help to temper some of the wilder forecasts about our supposedly AI-run future, utopian and dystopian alike.

[Brian C]

One of my chief bugbears is the term itself, "AI". It's now popularly used to describe a wide range of tools – from LLMs to generated images to older-generation stuff like chatbots and NPCs in videogames – that aren't necessarily doing the same thing, and in many cases are just standard computing dealing with bigger loads of data. As "AI", all of this gets placed under an umbrella term that evokes science-fiction concepts about sentient robots, none of which are remotely anything like what's being achieved with these technologies.

Very well said and I agree 100%.

Today I actually took it up on the offer to draft an email for me, since it was so excited about draft emails yesterday ... and it was a truly terrible, mostly off-topic email. I gave up after 3 attempts at revisions with more focused prompts, with the last prompt practically composing the email myself. But it just kept trying to go off in its own random directions and adding its own inappropriate embellishments. It's no closer to taking over my job - or any job - than a Furby.

[zedz]

Somebody once described “AI” to me as the evolution of predictive text. It looks like it’s writing, but it isn’t really, it’s just regurgitating what seems most likely to follow, based on bulk data. So while it can emulate generic content pretty well (e.g. a thank you letter), it isn’t much use as a creative tool, and totally unreliable as a source of information.

[cdnchris]

Yeah, it can't really "create", it just predicts the best response based on whatever data it has available (which can be small or huge), which it "learns" from various means, whether human mapped or through tools that allow it to "learn." It recognizes the patterns and matches. Part of the reason it has hallucinations is because along the way it picked up a bad match, assumes it's true, and continues on.

[MichaelB]

And some hallucinations can be downright libellous. As I encountered when Google AI made a somewhat eyebrow-raising allegation about someone who was very much alive.

The allegation in question should have been attached to a deceased colleague, but either the source was clunkily worded or the AI system parsed it incorrectly (or, of course, both), and…

…well, has there been an actual defamation case yet involving something generated by AI? Or is the “This may not be accurate” disclaimer considered sufficient?

[Adam X]

Val Kilmer to be digitally exhumed for upcoming film he was involved with, prior to his death.
Couldn’t quite decide where to put this, but thought I’d leave Kilmer’s ‘passages’ thread in peace.

I’m not sure how digitally recreating someone and imagining their ‘performance’ for a commercial film, can in any way be considered ethical, no matter how the filmmakers attempt to justify it. Curious how the rest of the cast from the already shot portion of the film feel about this.

[Matt]

People are discovering that the audiobook of Liza Minnelli's new autobiography has been recorded with an AI version of her voice. Last year, she and Lorna Luft licensed Judy Garland's voice to ElevenLabs so that you can hear one of the greatest entertainers of the 20th century read your PDFs, and then this year Liza released an AI-generated EDM single. It's speculated that Michael Feinstein, Liza's best friend and collaborator, recorded Liza telling her stories, edited them into a book, and then allowed the recordings to be used to generate audiobook narration. If you've watched any of Liza's recent appearances or interviews, you can see that she can barely get a full sentence out let alone narrate an 18-hour audiobook.

Anyway, the most damning evidence is a clip of her talking about her famous song "Liza with a Z," and I've been laughing about it all day. Poor Liza.

[Roger Ryan]

Val Kilmer to be digitally exhumed for upcoming film he was involved with, prior to his death.
Couldn’t quite decide where to put this, but thought I’d leave Kilmer’s ‘passages’ thread in peace.

I’m not sure how digitally recreating someone and imagining their ‘performance’ for a commercial film, can in any way be considered ethical, no matter how the filmmakers attempt to justify it. Curious how the rest of the cast from the already shot portion of the film feel about this.

The fact that the real Kilmer didn't shoot a frame of this film is what's most off-putting. This is not like trying to find a way to cover the loss of Oliver Reed in Gladiator or Brandon Lee in The Crow. The producers decided to not recast the role Kilmer was scheduled to play and to delete those scenes from the screenplay. Now they're saying the film doesn't work without those scenes, but there's no money to shoot additional footage (even though there's money to digitally create a fake Kilmer for the scenes?). Nope, it's a publicity gimmick and nothing else.

[Lowry_Sam]

Does Mick Jagger's skin care routine include a big dose of AI? He looks significantly younger than Donald Trump despite being 3 years older in the
In The Stars video.

Nevermind, found my answer with some help of AI (Google): South Park duo de-ages The Rolling Stones.

[Roger Ryan]

Does Mick Jagger's skin care routine include a big dose of AI? He looks significantly younger than Donald Trump despite being 3 years older in the
In The Stars video.

Nevermind, found my answer with some help of AI (Google): South Park duo de-ages The Rolling Stones.

Yeah, that's clearly one of the gimmicks of this video (to make Jagger, Richards, and Wood all appear as they did in the early 70s). I also don't think they used 20+ session guitarists, in addition to Richards and Wood, on this track nor did the track actually require dozens of drummers . I think the video is emulating Godard's footage of the band recording "Sympathy For The Devil" in '68 with the surreal twist of having too many folks in on the session.

[Lowry_Sam]

Yes, they do look like they did in the 70s, but clothes and set looks like the 60s. I didn't notice any glaring errors on the first view.

[Roger Ryan]

Yes, they do look like they did in the 70s, but clothes and set looks like the 60s. I didn't notice any glaring errors on the first view.

I know the article indicates the goal was to make the Stones look like they did in the "early 70s", but I really do think the video creator's intent was to have the studio setting and Stones appear as they did in Godard's footage shot in 1968 with everybody else appearing as modern day fans.